Amara Osei
Prompt Engineer & Workflow Writer
Treats prompting as an experiment, not an art: isolate one variable, run the batch, keep the receipts. Her prompt libraries ship with failure rates, not just the wins.
Beat: AI Video
- Prompt engineering & A/B testing
- Camera-language prompt libraries
- Workflow automation
- Failure-rate documentation
How Amara tests
One variable at a time, a fixed seed where possible, and a logged batch for every claim. My libraries publish the failure rate next to the recipe, because a prompt that works three times in ten needs to say so.
Latest from Amara
GuidesHow to Create an AI Avatar That Actually Passes QC
I ran 62 AI avatar attempts across four pipelines and logged the hit rate for each. Here is the reference-photo spec, the variants that failed, and real cost.
GuidesAI Captions: The Accuracy Numbers Nobody Publishes
Whisper Large-v3 sits near 4.2% word error rate on clean audio. Dirty audio drops tools to 70-85%. The fix is upstream of whichever tool you picked.
GuidesMidjourney Video: What It Actually Does, and What It Costs
Midjourney's video model animates a still you already made. No text-to-video, 480p output, and a credit cost around 8x an image job. Here's the real workflow.
Prompts40 Image-to-Video Motion Prompts That Won't Wreck Your Still
40 motion-only image-to-video prompts for Seedance, Kling and Veo — ambient motion, subject actions, camera drift, atmosphere — and where each one breaks.
Prompts30 Cinematic Lighting & Grade Prompts for AI Video Models
30 tested lighting and grade clauses for AI video — golden hour, low-key, practicals, film stocks — with which models actually read them and when to use each.